Despeckle ﬁltering applications has been a rapidly emerging research area in recent years. The basic
principles, the theoretical background, and the algorithmic steps of a representative set of despeckle
ﬁlters were covered in this book. Moreover, selected representative applications of image despeck-
ling covering a variety of ultrasound image processing tasks are presented. Most importantly, a
despeckle ﬁltering and evaluation protocol is documented in Table 6.1. The source code of the algo-
rithms discussed in this book has been made available on the web, thus enabling researchers to more
easily exploit the application of despeckle ﬁltering in their problems under investigation.
A total of 11 different despeckle ﬁlters were documented in this book based on linear ﬁl-
tering, nonlinear ﬁltering, diffusion ﬁltering, and wavelet ﬁltering. We have evaluated despeckle
ﬁltering on 440 (220 asymptomatic and 220 symptomatic) ultrasound images of the carotid artery
bifurcation, based on visual evaluation by two medical experts, texture analysis measures, and image
quality evaluation metrics. A linear despeckle ﬁlter based on local statistics (DsFlsmv) improved
the class separation between the asymptomatic and the symptomatic classes, gave only a marginal
improvement in the percentage of correct classiﬁcations success rate based on texture analysis and
the kNN classiﬁer, and improved the visual assessment by the experts. It was also found that the
despeckle ﬁlter can be used for despeckling asymptomatic images where the expert is
interested mainly in the plaque composition and texture analysis, whereas a geometric despeckle
ﬁlter (DsFgf4d ) can be used for despeckling of symptomatic images where the expert is interested
in identifying the degree of stenosis and the plaque borders. The results of this study suggest that
the ﬁrst-order statistics despeckle ﬁlter DsFlsmv may be applied on ultrasound images to improve
the visual perception and automatic image analysis.
Furthermore, despeckle ﬁltering was investigated as a preprocessing step for the automated
segmentation of the IMT  and the carotid plaque , followed by the carotid plaque texture
analysis, and classiﬁcation (as documented in the above paragraph). Despeckle ﬁlters DsFlsmv,
, and DsFgf4d gave the best performance for the segmentation tasks. It was shown in
Ref.  that when normalization and speckle reduction ﬁltering is applied on ultrasound images
of the carotid artery before IMT segmentation, the automated segmentation measurements are
C H A P T E R 6
Summary and Future Directions
138 DESPECKLE FILTERING ALGORITHMS
closer to the manual measurements. This ﬁeld has also been investigated by our group . Our
ﬁndings showed promising results; however, further work is required to evaluate the performance of
the suggested despeckle ﬁlters at a larger scale as well as their impact in clinical practice. In addi-
tion, the usefulness of the proposed despeckle ﬁlters, in portable ultrasound systems and in wireless
telemedicine systems still has to be investigated.
Our results on image quality evaluation (for comparing two different ultrasound scanners,
ATL HDI-3000 and ATL HDI-5000) showed that normalization and speckle-reduction ﬁltering
are important preprocessing steps favoring image quality. In addition, the usefulness of the proposed
TABLE 6.1 Despeckle Filtering and Evaluation Protocol
DESPECKLE FILTERING AND EVALUATION PROTOCOL
Recording of ultrasound images: Ultrasound images are acquired by ultrasound
equipment and stored for further image processing. Regions of interest (ROIs)
could be selected for further processing.
Normalize the image: The stored images may be retrieved, and a normalized
procedure may be applied (as described for example in Section 3.2).
Apply despeckle ﬁltering: Select the set of ﬁlters to apply despeckling together
with their corresponding parameters (like moving window size, iterations,
Texture features analysis: After despeckle ﬁltering, the user may select ROIs
(i.e., the plaque or the area around the IMC) and extract texture features.
Distance metrics between the original and the despeckled images may be
computed (as well as between different classes of images if applicable).
Compute image quality evaluation metrics: On the selected ROIs, compute
image quality evaluation metrics between the original noisy and the
Visual quality evaluation by experts: The original and/or despeckled images
may be visually evaluated by experts.
Select the most appropriate despeckle ﬁlter/ﬁlters: Based on steps 3 to 6,
construct a performance evaluation table (see for example Table 5.1) and
select the most appropriate ﬁlter(s) for the problem under investigation.